VP Ellipsis and Argument Structure Alternations in Muira Dargwa #
@cite{kalyakin-2026}
@cite{kalyakin-2026} argues that v-stranding VPE (vVPE) exists in Muira Dargwa (Nakh-Dagestanian) complex predicates: the light verb (= v) survives while its complement (= VP, containing the nominal root) is elided. The construction was first identified in Persian (@cite{toosarvandani-2009}); Muira Dargwa provides independent evidence and novel argument-structure alternation diagnostics.
Key Empirical Findings #
Causative alternation under vVPE: An inchoative antecedent can license a causative ellipsis site and vice versa — the same root with different Voice flavors. This is blocked in English VPE (@cite{merchant-2013}).
Antipassive blocking: Antipassive roots are coerced to v-adjunction (manner/activity position), placing them outside vVPE's deletion domain (VP). Antipassive roots therefore cannot be elided.
Again diagnostic: Under vVPE, BOTH repetitive and restitutive readings of again survive (@cite{kalyakin-2026} §4.1, exx. 52a–b). This contrasts with English VPE (only repetitive survives) and confirms the deletion domain is VP (complement of v), not vP. @cite{toosarvandani-2009} (ex. 90) independently shows both readings available for Persian vVPE.
Theoretical Analysis #
The analysis extends @cite{merchant-2013}'s [E]-feature theory: placing [E] on v (rather than Voice) yields a smaller deletion domain (VP rather than vP). This correctly predicts:
- Voice mismatches tolerated (same as English VPE)
- Transitivity/alternation mismatches tolerated (UNLIKE English VPE)
- Lexical verb mismatches still blocked (V is inside VP)
The causative alternation tolerance follows directly from existing Voice decomposition (@cite{kratzer-1996}, @cite{cuervo-2003}): alternating pairs share the same root [vGO, vBE]; only Voice differs. Since Voice is outside vVPE's deletion domain, mismatches in Voice are tolerated.
The core prediction: vVPE tolerates both voice and transitivity mismatches while blocking lexical verb mismatches.
English VPE has a strictly more restrictive profile than vVPE: it additionally blocks transitivity mismatches.
A datum for ellipsis under argument structure alternation.
- description : String
Description
- antecedentVoice : Minimalism.VoiceFlavor
Voice flavor of the antecedent
- targetVoice : Minimalism.VoiceFlavor
Voice flavor of the ellipsis site
- rootStructure : List Minimalism.VerbHead
Root event structure (shared between antecedent and target)
- grammatical : Bool
Is the alternation grammatical under the given ellipsis type?
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Inchoative antecedent → causative target (OK under vVPE). "The door opened. Then someone made it [open]." Voice: nonThematic → agentive; root: [vGO, vBE] shared.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Causative antecedent → inchoative target (OK under vVPE). "Someone opened the door, and then it [opened] by itself." Voice: agentive → nonThematic; root: [vGO, vBE] shared.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Same alternation is blocked in English VPE (@cite{merchant-2013}): v_trans ≠ v_unacc inside the deletion domain.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The shared root structure [vCAUSE, vGO, vBE] yields different decompositions under different Voice flavors — this is the causative alternation from Voice.lean.
The causative alternation is tolerated under vVPE because transitivity mismatches are allowed.
The causative alternation is blocked under English VPE because transitivity mismatches are blocked.
Bridge: each datum's grammaticality matches Merchant's canMismatch
prediction for the relevant ellipsis type and mismatch dimension.
The connection is structural: AlternationDatum.grammatical was set
to match the empirical judgment; canMismatch derives the same value
from spine positions.
Antipassive roots in Muira Dargwa are coerced to v-adjunction, placing them outside vVPE's deletion domain. They therefore cannot be elided under vVPE.
Change-of-state roots (object-adjoined) ARE inside vVPE's deletion domain — they can be elided.
The hierarchy of mismatch tolerance across ellipsis types: sluicing < English VPE < vVPE. Each step down tolerates strictly more mismatches.
vVPE's [E] position (v) is strictly below English VPE's (Voice). By monotonicity, any mismatch tolerated by English VPE is also tolerated by vVPE.
End-to-end chain: Voice severing (@cite{kratzer-1996}) → Merchant's deletion domain (@cite{merchant-2013}) → Kalyakin's empirical finding (@cite{kalyakin-2026}).
Step 1 (Voice.lean): The root [vCAUSE, vGO, vBE] yields a causative
decomposition [vDO, vCAUSE, vGO, vBE] under agentive Voice but an
inchoative [vCAUSE, vGO, vBE] under nonThematic Voice. The full
decompositions differ,
but the root (= VP content) is shared.
Step 2 (DeletionDomain.lean): Under vVPE ([E] on v), the deletion domain is VP. Transitivity (determined by v) is external to VP, so transitivity mismatches are tolerated.
Step 3 (this file): The datum — inchoative→causative alternation under vVPE is grammatical — matches the prediction.
Convergent prediction: Merchant's theory correctly predicts that sluicing (C[E]) blocks voice mismatches — Voice is inside TP, the deletion domain of sluicing. The SCSS corpus (@cite{anand-hardt-mccloskey-2021}, §5.5) independently confirms this: across 4,700 annotated sluices, zero antecedent–ellipsis site pairings exhibit active/passive voice mismatches. The same theoretical apparatus that Kalyakin extends to vVPE already works for sluicing.
Under vVPE, BOTH repetitive and restitutive again survive. This contrasts with English VPE (only repetitive survives) and confirms the deletion domain is VP (complement of v), not vP.
@cite{kalyakin-2026} §4.1 (exx. 52a–b): both repetitive and restitutive ʔibrra 'again' are available under vVPE in Muira Dargwa. @cite{toosarvandani-2009} (ex. 90) independently shows both readings available for Persian vVPE.
English VPE deletes restitutive again: only repetitive survives. (@cite{merchant-2013}, building on Johnson 2004, von Stechow 1996).
The again contrast directly distinguishes vVPE from English VPE: same test, different result — proving different deletion domains.
Is a CPr's NV inside vVPE's deletion domain?
The fragment's AnnotatedCPr stores RootPosition (from
Core.Lexical.LevinClass); this function bridges to the
Minimalist rootInVVPEDomain from DeletionDomain.lean.
Equations
Instances For
Change-of-state NVs (complement position) are inside vVPE's deletion domain: they can be elided.
Manner/activity NVs (adjoined position) are outside vVPE's deletion domain: they survive ellipsis. This is why antipassive roots (coerced to adjunction) block vVPE.
Datum for the NV-drop constituency test. @cite{kalyakin-2026} §3.2 distinguishes vVPE from argument ellipsis (AE):
- vVPE: NV+argument deleted together (constituent = VP)
- AE: argument alone deleted, NV survives
- *NV alone deleted, argument survives → ungrammatical The ungrammaticality of NV-only deletion proves the elided constituent is VP (containing both NV and argument), not just NV.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
NV + argument both dropped (= vVPE): grammatical.
Equations
Instances For
Argument alone dropped (= argument ellipsis): grammatical.
Equations
Instances For
NV alone dropped, argument survives: ungrammatical. This rules out NV-drop as a process distinct from vVPE — you can't delete just the NV without its complement.
Equations
Instances For
The NV-drop test confirms constituent deletion: NV+arg is a constituent (VP), NV alone is not.
Bridge: Muira Dargwa, Persian, British English share [E] on v (= vVPE), while Bangla's deletion domain is vP (= English VPE with LV evacuation via head movement).
Muira Dargwa and British do pattern together: same [E] position, no VIR, arg-structure alternations tolerated.
Persian has same [E] position as Muira Dargwa but stricter identity: VIR blocks arg-structure alternations in Persian that Dargwa allows.